Meta has expanded its custom silicon partnership with Broadcom, a move that says a great deal about where the company sees the next phase of AI competition. In announcements, Meta said Broadcom will help co-develop multiple generations of its MTIA chips, the custom accelerators Meta uses for AI across its apps and services. The expanded agreement runs through 2029 and includes an initial commitment of more than 1 gigawatt of computing capacity, with Broadcom CEO Hock Tan moving from Meta’s board into an advisory role on chip strategy.
“Meta is partnering with Broadcom across chip design, packaging, and networking to build out the massive computing foundation we need to deliver personal superintelligence to billions of people,” said Meta founder and CEO Mark Zuckerberg. “As we roll out more than 1GW of our custom silicon to start and then multiple gigawatts over time, this Meta and Broadcom partnership will give us greater performance and efficiency for everything we’re building.”
Meta is building a long-term hardware program designed to support ranking, recommendations, inference, and eventually more demanding AI workloads. Meta said the first MTIA chip, MTIA 300, is already in use, three more chips are due through 2027, with the company describing the new roadmap as part of a broader push to expand its data center footprint.
According to the Meta and Broadcom partnership agreement, the companies will work together across chip design, advanced packaging, and networking, with Broadcom’s Ethernet hardware used to connect Meta’s AI clusters. That is a crucial detail. At this level, AI infrastructure is not only about raw chip performance; it is also about how efficiently data can move across racks, nodes, and entire data centers. Meta and Broadcom are signaling that the bottleneck is now as much system design as it is silicon design.
This deal also fits into Meta’s broader spending pattern. According to Reuters, in March, Meta expects capital spending of between $115 billion and $135 billion this year, and in February, the company agreed to buy up to $60 billion worth of AI chips from AMD over five years. That followed other large chip commitments, showing that Meta is spreading its bets across suppliers instead of relying on a single hardware stack. The strategy is straightforward: secure enough compute, reduce vendor dependence, and keep pace with the cost of running large AI systems at scale.
The big technology firms are designing their own chips to reduce dependence on Nvidia’s expensive processors, and Broadcom has become one of the biggest beneficiaries of that shift. Just days earlier, Broadcom had signed a long-term deal with Google to develop and supply future generations of custom AI chips through 2031. Taken together, the Google and Meta deals suggest Broadcom is becoming a key contractor in the custom-chip era, not just a component supplier.
The Hock Tan board transition is a smaller detail, but it is still worth noting. Meta said he will move into an advisory role, where he will continue to provide guidance on Meta’s custom silicon roadmap and infrastructure investments. In practical terms, that suggests Meta wants continuity and technical input without keeping a Broadcom executive on the board while the partnership grows in scope. It is a governance change, but it also reads like a sign of how deeply the two companies are now tied together on AI infrastructure.
Meta is treating AI infrastructure as a strategic asset, not a support function. The company’s public language about “personal superintelligence” is ambitious, but the operational message is simpler: it wants more control over performance, cost, and supply. The one-gigawatt figure is best understood as the first phase of a much larger rollout, not a finished build. In a market where compute is scarce and expensive, that is the kind of commitment that can shape both Meta’s product roadmap and its relationship with the broader chip industry.
[Related Post: AMD and Meta Expand Multiyear AI Infrastructure Partnership ]
